Development and assessment of a 250 m spatial resolution MODIS annual land cover time series (2000–2011) for the forest region of Canada derived from change-based updating Darren Pouliot a, ⁎, Rasim Latifovic a , Natalie Zabcic a , Luc Guindon b , Ian Olthof a a Natural Resources Canada, Earth Sciences Sector, Canada Centre for Remote Sensing, 588 Booth Street, Ottawa, Ontario K1A0Y7, Canada b Natural Resources Canada, Canadian Forest Service, Laurentian Forestry Centre, Sainte-Foy, Quebec, Canada abstract article info Article history: Received 10 April 2013 Received in revised form 4 October 2013 Accepted 6 October 2013 Available online 4 November 2013 Keywords: Land cover Time series MODIS Change detection Boreal Accuracy Detailed information on the spatial and temporal distribution of land cover is required to evaluate the effects of land cover change on environmental processes. The development of temporally consistent land cover time series (LCTS) from satellite-based earth observation has proven difficult because multi-year observations are acquired under different conditions resulting in high inter-annual reflectance variability. This leads to spurious differences in land cover when standard approaches for image classification are applied to generate multi-year land cover data. To reduce this effect, a common solution has been to first detect change and update a base map for only these change areas. As long as the change commission error is low, this approach will ensure high consistency between maps in the time series. Here we present an approach for change-based LCTS development following from previous research, but with significant advancements in change detection, training, classification, and evidence-based refinement. The method was applied to generate an annual LCTS covering Canada spanning 2000–2011 that is consistent between years and can be used to identify dominant change transitions. Assessment of the LCTS was challenging because multiple maps needed to be evaluated and can be prohibitive particularly for annual time series covering several years. Three approaches were undertaken involving visual examination, comparison with a reference sample derived from Landsat, and comparison with the MODIS Global LCTS V5.1. Visual assessment revealed high inter-map consistency and logical temporal change trajectories of land cover classes. Comparison with the reference sample showed an accuracy of 70% at the 19 class thematic resolution. Accounting for mixed pixels by considering the first or second reference land cover label as correct increased the accuracy to 80%. Comparison with the MODIS Global LCTS showed that the Canada LCTS achieved higher inter-map consistency and accuracy as expected with national relative to global land cover products. Crown Copyright © 2013 Published by Elsevier Inc. All rights reserved. 1. Introduction Land cover and land cover change is known to affect the environment in ways that can impact human health by altering climate, weather, water, air, biodiversity, wildlife, disease risk, and food security (Chhabra et al., 2006). To better quantify and potentially mitigate the undesirable effects of land cover change requires spatially and temporally extensive information so that linkages between land cover and ecosystem properties can be identified. Satellite earth observation and extraction algorithms have made land cover information more widely available, but it is often static representing only one point in time. This is largely due to the difficulty of efficiently creating two or more land cover maps where the occurrence of false change between them is small (Bontemps et al., 2012; Defourny & Bontemps, 2012; Pouliot, Latifovic, Olthof, & Fraser, 2012). Methods for accurate detection of more general change/no-change classes have been developed and evaluated for large regional applications (Bucha & Stibig, 2008; Fraser, Abuelgasim, & Latifovic, 2005; Hansen et al., 2008; Masek et al., 2008; Potapov, Hansen, Stehman, Loveland, & Pittman, 2008; Pouliot, Latifovic, Fernandes, & Olthof, 2009; Zhan et al., 2002). However, for many applications more detailed information regarding the land cover class before and after change is needed (Ramankutty et al., 2006). Comparing maps between periods, known as post-classification comparison, is one approach to deriving this type of from–to change information. As a general rule, the accuracy of change detection based on post-classification comparison is the product of the map accuracies (Stow, Tinney, & Estes, 1980). Currently, for 30 m resolution maps, an accepted accuracy is ~85% (Foody, 2002). However, for large area vegetation classification lower accuracies are typically achieved (Franklin, Lavigne, Wulder, & Stenhouse, 2002). At large regional scales using coarser spatial resolution data (250 m–500 m), reported accuracies range from 67%–75% (Bontemps et al., 2012; Friedl et al., 2010; Latifovic et al., 2012). Taking the upper bound of 75% as the Remote Sensing of Environment 140 (2014) 731–743 ⁎ Corresponding author at: Natural Resources Canada, Earth Sciences Sector, Canada Centre for Remote Sensing, 588 Booth Street, Ottawa, Ontario, K1A0Y7, Canada. Tel.: +1 613 947 1267. E-mail address: Darren.pouliot@ccrs.nrcan.gc.ca (D. Pouliot). 0034-4257/$ – see front matter. Crown Copyright © 2013 Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.rse.2013.10.004 Contents lists available at ScienceDirect Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse